Multiclass Feature Selection Model for Adversarial Attacks in IoT Environment

Malware attacks have become prevalent in the uti-lization of Internet of Things (IoT) networks. The security and re-liability of IoT networks and information systems are significantly threatened by this. Machine learning-based Intrusion Detection Systems (IDS) are ineffective in countering these cyber threats due to the widespread distribution of malware. This research presents a method for choosing features that utilize Generative Adversarial Networks (GAN) to enhance the effectiveness of Intrusion Detection Systems (IDS) by reliably and promptly identifying both new and previously known threats in real-time in the IoT environment. Our feature selection method consists of two parts. First, we employed the Mutual Information (MI) technique to reduce the required features. Second, we employed exclusion and inclusion feature selection techniques which rely on discriminator accuracy to detect adversarial attacks. We performed an experiment using the proposed method on the CICIOT 2023 dataset, which includes a diverse range of IoT attack incidents. To the best of our knowledge, our proposed effort is the initial endeavor to utilize a feature selection method on this recently introduced dataset. The selection of 20 features for generating new data with optimal evaluation accuracy was accomplished using a Convolutional Neural Network (CNN) and a Mutual Information Classifier, incorporating exclusion and inclusion techniques. Ultimately, we utilized a Recurrent Neural Network (RNN) with a limited set of features to classify malware attacks. Subsequently, we evaluated the correctness of the classification by considering prominent features, resulting in a commendable accuracy rate of 93%.